4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
36 #include <linux/kobject.h>
37 #include <linux/spinlock.h>
38 #include <linux/sysfs.h>
40 #include <uapi/linux/lustre/lustre_idl.h>
41 #include <lustre_lib.h>
42 #include <libcfs/bitmap.h>
43 #ifdef HAVE_SERVER_SUPPORT
44 # include <lu_target.h>
45 # include <obd_target.h>
48 #include <lustre_export.h>
49 #include <lustre_fid.h>
50 #include <lustre_fld.h>
51 #include <lustre_handles.h>
52 #include <lustre_intent.h>
54 #include <lustre_quota.h>
56 #define MAX_OBD_DEVICES 8192
64 struct lov_oinfo { /* per-stripe data structure */
65 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
66 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
67 int loi_ost_gen; /* generation of this loi_ost_idx */
69 unsigned long loi_kms_valid:1;
70 __u64 loi_kms; /* known minimum size */
71 struct ost_lvb loi_lvb;
72 struct osc_async_rc loi_ar;
75 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
78 oinfo->loi_kms_valid = 1;
84 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
86 /* obd info for a particular level (lov, osc). */
88 /* OBD_STATFS_* flags */
90 /* statfs data specific for every OSC, if needed at all. */
91 struct obd_statfs *oi_osfs;
92 /* An update callback which is called to update some data on upper
93 * level. E.g. it is used for update lsm->lsm_oinfo at every received
94 * request in osc level for enqueue requests. It is also possible to
95 * update some caller data from LOV layer if needed. */
96 obd_enqueue_update_f oi_cb_up;
100 struct list_head typ_chain;
101 struct obd_ops *typ_dt_ops;
102 struct md_ops *typ_md_ops;
103 struct proc_dir_entry *typ_procroot;
104 struct proc_dir_entry *typ_procsym;
105 __u32 typ_sym_filter;
108 struct lu_device_type *typ_lu;
109 spinlock_t obd_type_lock;
110 struct kobject typ_kobj;
111 struct completion typ_kobj_unregister;
121 struct timeout_item {
122 enum timeout_event ti_event;
126 struct list_head ti_obd_list;
127 struct list_head ti_chain;
130 #define OBD_MAX_RIF_DEFAULT 8
131 #define OBD_MAX_RIF_MAX 512
132 #define OSC_MAX_RIF_MAX 256
133 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
134 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
135 #define OSC_DEFAULT_RESENDS 10
137 /* possible values for fo_sync_lock_cancel */
139 NEVER_SYNC_ON_CANCEL = 0,
140 BLOCKING_SYNC_ON_CANCEL = 1,
141 ALWAYS_SYNC_ON_CANCEL = 2,
142 NUM_SYNC_ON_CANCEL_STATES
146 * Limit reply buffer size for striping data to one x86_64 page. This
147 * value is chosen to fit the striping data for common use cases while
148 * staying well below the limit at which the buffer must be backed by
149 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
152 #define OBD_MAX_DEFAULT_EA_SIZE 4096
154 enum obd_cl_sem_lock_class {
163 struct rw_semaphore cl_sem;
164 struct obd_uuid cl_target_uuid;
165 struct obd_import *cl_import; /* ptlrpc connection state */
166 size_t cl_conn_count;
168 /* Cache maximum and default values for easize. This is
169 * strictly a performance optimization to minimize calls to
170 * obd_size_diskmd(). The default values are used to calculate the
171 * initial size of a request buffer. The ptlrpc layer will resize the
172 * buffer as needed to accommodate a larger reply from the
173 * server. The default values should be small enough to avoid wasted
174 * memory and excessive use of vmalloc(), yet large enough to avoid
175 * reallocating the buffer in the common use case. */
177 /* Default EA size for striping attributes. It is initialized at
178 * mount-time based on the default stripe width of the filesystem,
179 * then it tracks the largest observed EA size advertised by
180 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
181 __u32 cl_default_mds_easize;
183 /* Maximum possible EA size computed at mount-time based on
184 * the number of OSTs in the filesystem. May be increased at
185 * run-time if a larger observed size is advertised by the MDT. */
186 __u32 cl_max_mds_easize;
188 enum lustre_sec_part cl_sp_me;
189 enum lustre_sec_part cl_sp_to;
190 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
192 /* the grant values are protected by loi_list_lock below */
193 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
194 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
195 unsigned long cl_dirty_transit; /* dirty synchronous */
196 unsigned long cl_avail_grant; /* bytes of credit for ost */
197 unsigned long cl_lost_grant; /* lost credits (trunc) */
198 /* grant consumed for dirty pages */
199 unsigned long cl_dirty_grant;
201 /* since we allocate grant by blocks, we don't know how many grant will
202 * be used to add a page into cache. As a solution, we reserve maximum
203 * grant before trying to dirty a page and unreserve the rest.
204 * See osc_{reserve|unreserve}_grant for details. */
205 long cl_reserved_grant;
206 struct list_head cl_cache_waiters; /* waiting for cache/grant */
207 time64_t cl_next_shrink_grant; /* seconds */
208 struct list_head cl_grant_shrink_list; /* Timeout event list */
209 time64_t cl_grant_shrink_interval; /* seconds */
211 /* A chunk is an optimal size used by osc_extent to determine
212 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
214 /* extent insertion metadata overhead to be accounted in grant,
216 unsigned int cl_grant_extent_tax;
217 /* maximum extent size, in number of pages */
218 unsigned int cl_max_extent_pages;
220 /* keep track of objects that have lois that contain pages which
221 * have been queued for async brw. this lock also protects the
222 * lists of osc_client_pages that hang off of the loi */
224 * ->cl_loi_list_lock protects consistency of
225 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
226 * ->ap_completion() call-backs are executed under this lock. As we
227 * cannot guarantee that these call-backs never block on all platforms
228 * (as a matter of fact they do block on Mac OS X), type of
229 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
230 * and blocking mutex on Mac OS X. (Alternative is to make this lock
231 * blocking everywhere, but we don't want to slow down fast-path of
232 * our main platform.)
234 * NB by Jinshan: though field names are still _loi_, but actually
235 * osc_object{}s are in the list.
237 spinlock_t cl_loi_list_lock;
238 struct list_head cl_loi_ready_list;
239 struct list_head cl_loi_hp_ready_list;
240 struct list_head cl_loi_write_list;
241 struct list_head cl_loi_read_list;
242 __u32 cl_r_in_flight;
243 __u32 cl_w_in_flight;
244 /* just a sum of the loi/lop pending numbers to be exported by /proc */
245 atomic_t cl_pending_w_pages;
246 atomic_t cl_pending_r_pages;
247 __u32 cl_max_pages_per_rpc;
248 __u32 cl_max_rpcs_in_flight;
249 struct obd_histogram cl_read_rpc_hist;
250 struct obd_histogram cl_write_rpc_hist;
251 struct obd_histogram cl_read_page_hist;
252 struct obd_histogram cl_write_page_hist;
253 struct obd_histogram cl_read_offset_hist;
254 struct obd_histogram cl_write_offset_hist;
256 /** LRU for osc caching pages */
257 struct cl_client_cache *cl_cache;
258 /** member of cl_cache->ccc_lru */
259 struct list_head cl_lru_osc;
260 /** # of available LRU slots left in the per-OSC cache.
261 * Available LRU slots are shared by all OSCs of the same file system,
262 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
263 atomic_long_t *cl_lru_left;
264 /** # of busy LRU pages. A page is considered busy if it's in writeback
265 * queue, or in transfer. Busy pages can't be discarded so they are not
267 atomic_long_t cl_lru_busy;
268 /** # of LRU pages in the cache for this client_obd */
269 atomic_long_t cl_lru_in_list;
270 /** # of threads are shrinking LRU cache. To avoid contention, it's not
271 * allowed to have multiple threads shrinking LRU cache. */
272 atomic_t cl_lru_shrinkers;
273 /** The time when this LRU cache was last used. */
274 time64_t cl_lru_last_used;
275 /** stats: how many reclaims have happened for this client_obd.
276 * reclaim and shrink - shrink is async, voluntarily rebalancing;
277 * reclaim is sync, initiated by IO thread when the LRU slots are
279 __u64 cl_lru_reclaim;
280 /** List of LRU pages for this client_obd */
281 struct list_head cl_lru_list;
282 /** Lock for LRU page list */
283 spinlock_t cl_lru_list_lock;
284 /** # of unstable pages in this client_obd.
285 * An unstable page is a page state that WRITE RPC has finished but
286 * the transaction has NOT yet committed. */
287 atomic_long_t cl_unstable_count;
288 /** Link to osc_shrinker_list */
289 struct list_head cl_shrink_list;
291 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
292 atomic_t cl_destroy_in_flight;
293 wait_queue_head_t cl_destroy_waitq;
295 struct mdc_rpc_lock *cl_rpc_lock;
297 /* modify rpcs in flight
298 * currently used for metadata only */
299 spinlock_t cl_mod_rpcs_lock;
300 __u16 cl_max_mod_rpcs_in_flight;
301 __u16 cl_mod_rpcs_in_flight;
302 __u16 cl_close_rpcs_in_flight;
303 wait_queue_head_t cl_mod_rpcs_waitq;
304 unsigned long *cl_mod_tag_bitmap;
305 struct obd_histogram cl_mod_rpcs_hist;
308 struct mutex cl_mgc_mutex;
309 struct local_oid_storage *cl_mgc_los;
310 struct dt_object *cl_mgc_configs_dir;
311 atomic_t cl_mgc_refcount;
312 struct obd_export *cl_mgc_mgsexp;
314 /* checksumming for data sent over the network */
315 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
316 cl_checksum_dump:1; /* same */
317 /* supported checksum types that are worked out at connect time */
318 __u32 cl_supp_cksum_types;
319 /* checksum algorithm to be used */
320 enum cksum_types cl_cksum_type;
322 /* also protected by the poorly named _loi_list_lock lock above */
323 struct osc_async_rc cl_ar;
325 /* sequence manager */
326 struct lu_client_seq *cl_seq;
327 struct rw_semaphore cl_seq_rwsem;
329 atomic_t cl_resends; /* resend count */
331 /* ptlrpc work for writeback in ptlrpcd context */
332 void *cl_writeback_work;
334 /* hash tables for osc_quota_info */
335 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
336 /* Links to the global list of registered changelog devices */
337 struct list_head cl_chg_dev_linkage;
339 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
346 struct echo_client_obd {
347 struct obd_export *ec_exp; /* the local connection to osc/lov */
349 struct list_head ec_objects;
350 struct list_head ec_locks;
354 /* Generic subset of OSTs */
356 __u32 *op_array; /* array of index of
358 unsigned int op_count; /* number of OSTs in the array */
359 unsigned int op_size; /* allocated size of lp_array */
360 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
363 /* allow statfs data caching for 1 second */
364 #define OBD_STATFS_CACHE_SECONDS 1
366 struct lov_tgt_desc {
367 struct list_head ltd_kill;
368 struct obd_uuid ltd_uuid;
369 struct obd_device *ltd_obd;
370 struct obd_export *ltd_exp;
372 __u32 ltd_index; /* index in lov_obd->tgts */
373 unsigned long ltd_active:1,/* is this target up for requests */
374 ltd_activate:1,/* should target be activated */
375 ltd_reap:1; /* should this target be deleted */
379 struct lov_desc desc;
380 struct lov_tgt_desc **lov_tgts; /* sparse array */
381 struct ost_pool lov_packed; /* all OSTs in a packed
383 struct mutex lov_lock;
384 struct obd_connect_data lov_ocd;
385 struct proc_dir_entry *targets_proc_entry;
386 atomic_t lov_refcount;
387 __u32 lov_death_row; /* tgts scheduled to be deleted */
388 __u32 lov_tgt_size; /* size of tgts array */
391 struct cfs_hash *lov_pools_hash_body; /* used for key access */
392 struct list_head lov_pool_list; /* used for sequential access */
393 struct proc_dir_entry *lov_pool_proc_entry;
394 enum lustre_sec_part lov_sp_me;
396 /* Cached LRU and unstable data from upper layer */
397 struct cl_client_cache *lov_cache;
399 struct rw_semaphore lov_notify_lock;
402 struct lmv_tgt_desc {
403 struct obd_uuid ltd_uuid;
404 struct obd_export *ltd_exp;
406 struct mutex ltd_fid_mutex;
407 unsigned long ltd_active:1; /* target up for requests */
411 struct lu_client_fld lmv_fld;
413 struct lmv_desc desc;
414 struct obd_uuid cluuid;
415 struct proc_dir_entry *targets_proc_entry;
417 struct mutex lmv_init_mutex;
422 __u32 tgts_size; /* size of tgts array */
423 struct lmv_tgt_desc **tgts;
425 struct obd_connect_data conn_data;
428 struct niobuf_local {
429 __u64 lnb_file_offset;
430 __u32 lnb_page_offset;
434 struct page *lnb_page;
438 struct tgt_thread_big_cache {
439 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
442 #define LUSTRE_FLD_NAME "fld"
443 #define LUSTRE_SEQ_NAME "seq"
445 #define LUSTRE_MDD_NAME "mdd"
446 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
447 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
448 #define LUSTRE_VVP_NAME "vvp"
449 #define LUSTRE_LMV_NAME "lmv"
450 #define LUSTRE_SLP_NAME "slp"
451 #define LUSTRE_LOD_NAME "lod"
452 #define LUSTRE_OSP_NAME "osp"
453 #define LUSTRE_LWP_NAME "lwp"
455 /* obd device type names */
456 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
457 #define LUSTRE_MDS_NAME "mds"
458 #define LUSTRE_MDT_NAME "mdt"
459 #define LUSTRE_MDC_NAME "mdc"
460 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
461 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
462 #define LUSTRE_OSC_NAME "osc"
463 #define LUSTRE_LOV_NAME "lov"
464 #define LUSTRE_MGS_NAME "mgs"
465 #define LUSTRE_MGC_NAME "mgc"
467 #define LUSTRE_ECHO_NAME "obdecho"
468 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
469 #define LUSTRE_QMT_NAME "qmt"
471 /* Constant obd names (post-rename) */
472 #define LUSTRE_MDS_OBDNAME "MDS"
473 #define LUSTRE_OSS_OBDNAME "OSS"
474 #define LUSTRE_MGS_OBDNAME "MGS"
475 #define LUSTRE_MGC_OBDNAME "MGC"
477 static inline int is_lwp_on_mdt(char *name)
481 ptr = strrchr(name, '-');
483 CERROR("%s is not a obdname\n", name);
487 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
489 if (strncmp(ptr + 1, "MDT", 3) != 0)
492 while (*(--ptr) != '-' && ptr != name);
497 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
503 static inline int is_lwp_on_ost(char *name)
507 ptr = strrchr(name, '-');
509 CERROR("%s is not a obdname\n", name);
513 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
515 if (strncmp(ptr + 1, "OST", 3) != 0)
518 while (*(--ptr) != '-' && ptr != name);
523 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
530 * Events signalled through obd_notify() upcall-chain.
532 enum obd_notify_event {
533 /* Device connect start */
535 /* Device activated */
537 /* Device deactivated */
539 /* Connect data for import were changed */
541 /* Administratively deactivate/activate event */
542 OBD_NOTIFY_DEACTIVATE,
547 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
548 * being main example).
550 struct obd_notify_upcall {
551 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
552 enum obd_notify_event ev, void *owner);
553 /* Opaque datum supplied by upper layer listener */
557 struct target_recovery_data {
558 svc_handler_t trd_recovery_handler;
559 pid_t trd_processing_task;
560 struct completion trd_starting;
561 struct completion trd_finishing;
564 struct obd_llog_group {
565 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
566 wait_queue_head_t olg_waitq;
568 struct mutex olg_cat_processing;
571 /* corresponds to one of the obd's */
572 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
575 struct obd_type *obd_type;
576 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
577 int obd_minor; /* device number: lctl dl */
578 struct lu_device *obd_lu_dev;
580 /* common and UUID name of this device */
581 struct obd_uuid obd_uuid;
582 char obd_name[MAX_OBD_NAME];
584 /* bitfield modification is protected by obd_dev_lock */
586 obd_attached:1, /* finished attach */
587 obd_set_up:1, /* finished setup */
588 obd_recovering:1, /* there are recoverable clients */
589 obd_abort_recovery:1, /* recovery expired */
590 obd_version_recov:1, /* obd uses version checking */
591 obd_replayable:1, /* recovery enabled; inform clients */
592 obd_no_transno:1, /* no committed-transno notification */
593 obd_no_recov:1, /* fail instead of retry messages */
594 obd_stopping:1, /* started cleanup */
595 obd_starting:1, /* started setup */
596 obd_force:1, /* cleanup with > 0 obd refcount */
597 obd_fail:1, /* cleanup with failover */
598 obd_no_conn:1, /* deny new connections */
599 obd_inactive:1, /* device active/inactive
600 * (for /proc/status only!!) */
601 obd_no_ir:1, /* no imperative recovery. */
602 obd_process_conf:1, /* device is processing mgs config */
603 obd_uses_nid_stats:1, /* maintain per-client OBD stats */
604 obd_checksum_dump:1; /* dump pages upon cksum error */
606 /* use separate field as it is set in interrupt to don't mess with
607 * protection of other bits using _bh lock */
608 unsigned long obd_recovery_expired:1;
609 /* uuid-export hash body */
610 struct cfs_hash *obd_uuid_hash;
611 /* nid-export hash body */
612 struct cfs_hash *obd_nid_hash;
614 struct cfs_hash *obd_nid_stats_hash;
615 /* client_generation-export hash body */
616 struct cfs_hash *obd_gen_hash;
617 struct list_head obd_nid_stats;
618 struct list_head obd_exports;
619 struct list_head obd_unlinked_exports;
620 struct list_head obd_delayed_exports;
621 struct list_head obd_lwp_list;
622 atomic_t obd_refcount;
624 spinlock_t obd_nid_lock;
625 struct ldlm_namespace *obd_namespace;
626 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
627 /* a spinlock is OK for what we do now, may need a semaphore later */
628 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
629 spinlock_t obd_osfs_lock;
630 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
632 __u64 obd_last_committed;
633 struct mutex obd_dev_mutex;
634 struct lvfs_run_ctxt obd_lvfs_ctxt;
635 struct obd_llog_group obd_olg; /* default llog group */
636 struct obd_device *obd_observer;
637 struct rw_semaphore obd_observer_link_sem;
638 struct obd_notify_upcall obd_upcall;
639 struct obd_export *obd_self_export;
640 struct obd_export *obd_lwp_export;
641 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
642 struct list_head obd_exports_timed;
643 time64_t obd_eviction_timer; /* for ping evictor */
645 int obd_max_recoverable_clients;
646 atomic_t obd_connected_clients;
647 int obd_stale_clients;
648 /* this lock protects all recovery list_heads, timer and
649 * obd_next_recovery_transno value */
650 spinlock_t obd_recovery_task_lock;
651 __u64 obd_next_recovery_transno;
652 int obd_replayed_requests;
653 int obd_requests_queued_for_recovery;
654 wait_queue_head_t obd_next_transno_waitq;
655 /* protected by obd_recovery_task_lock */
656 struct timer_list obd_recovery_timer;
658 time64_t obd_recovery_start;
659 /* seconds, for lprocfs_status */
660 time64_t obd_recovery_end;
661 time64_t obd_recovery_time_hard;
662 time64_t obd_recovery_timeout;
663 int obd_recovery_ir_factor;
665 /* new recovery stuff from CMD2 */
666 int obd_replayed_locks;
667 atomic_t obd_req_replay_clients;
668 atomic_t obd_lock_replay_clients;
669 struct target_recovery_data obd_recovery_data;
671 /* all lists are protected by obd_recovery_task_lock */
672 struct list_head obd_req_replay_queue;
673 struct list_head obd_lock_replay_queue;
674 struct list_head obd_final_req_queue;
677 #ifdef HAVE_SERVER_SUPPORT
678 struct obd_device_target obt;
679 struct filter_obd filter;
681 struct echo_obd echo;
683 struct client_obd cli;
684 struct echo_client_obd echo_client;
689 /* Fields used by LProcFS */
690 struct lprocfs_stats *obd_stats;
691 unsigned int obd_cntr_base;
693 unsigned int obd_md_cntr_base;
694 struct lprocfs_stats *obd_md_stats;
696 struct proc_dir_entry *obd_proc_entry;
697 struct proc_dir_entry *obd_proc_exports_entry;
698 struct proc_dir_entry *obd_svc_procroot;
699 struct lprocfs_stats *obd_svc_stats;
700 struct attribute_group obd_attrs_group;
701 struct attribute **obd_attrs;
702 struct lprocfs_vars *obd_vars;
703 atomic_t obd_evict_inprogress;
704 wait_queue_head_t obd_evict_inprogress_waitq;
705 struct list_head obd_evict_list; /* protected with pet_lock */
708 * LDLM pool part. Save last calculated SLV and Limit.
710 rwlock_t obd_pool_lock;
714 int obd_conn_inprogress;
717 * List of outstanding class_incref()'s fo this OBD. For debugging. */
718 struct lu_ref obd_reference;
720 struct kset obd_kset; /* sysfs object collection */
721 struct kobj_type obd_ktype;
722 struct completion obd_kobj_unregister;
725 /* get/set_info keys */
726 #define KEY_ASYNC "async"
727 #define KEY_CHANGELOG_CLEAR "changelog_clear"
728 #define KEY_FID2PATH "fid2path"
729 #define KEY_CHECKSUM "checksum"
730 #define KEY_CLEAR_FS "clear_fs"
731 #define KEY_CONN_DATA "conn_data"
732 #define KEY_EVICT_BY_NID "evict_by_nid"
733 #define KEY_FIEMAP "fiemap"
734 #define KEY_FLUSH_CTX "flush_ctx"
735 #define KEY_GRANT_SHRINK "grant_shrink"
736 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
737 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
738 #define KEY_INTERMDS "inter_mds"
739 #define KEY_LAST_ID "last_id"
740 #define KEY_LAST_FID "last_fid"
741 #define KEY_MAX_EASIZE "max_easize"
742 #define KEY_DEFAULT_EASIZE "default_easize"
743 #define KEY_MGSSEC "mgssec"
744 #define KEY_READ_ONLY "read-only"
745 #define KEY_REGISTER_TARGET "register_target"
746 #define KEY_SET_FS "set_fs"
747 #define KEY_TGT_COUNT "tgt_count"
748 /* KEY_SET_INFO in lustre_idl.h */
749 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
751 #define KEY_CACHE_SET "cache_set"
752 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
753 #define KEY_OSP_CONNECTED "osp_connected"
757 static inline int it_to_lock_mode(struct lookup_intent *it)
759 /* CREAT needs to be tested before open (both could be set) */
760 if (it->it_op & IT_CREAT)
762 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
765 else if (it->it_op & IT_READDIR)
767 else if (it->it_op & IT_GETXATTR)
769 else if (it->it_op & IT_SETXATTR)
772 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
777 MF_MDC_CANCEL_FID1 = 1 << 0,
778 MF_MDC_CANCEL_FID2 = 1 << 1,
779 MF_MDC_CANCEL_FID3 = 1 << 2,
780 MF_MDC_CANCEL_FID4 = 1 << 3,
781 MF_GET_MDT_IDX = 1 << 4,
785 CLI_SET_MEA = 1 << 0,
786 CLI_RM_ENTRY = 1 << 1,
789 CLI_MIGRATE = 1 << 4,
793 * GETXATTR is not included as only a couple of fields in the reply body
794 * is filled, but not FID which is needed for common intent handling in
795 * mdc_finish_intent_lock()
797 static inline bool it_has_reply_body(const struct lookup_intent *it)
799 return it->it_op & (IT_OPEN | IT_UNLINK | IT_LOOKUP | IT_GETATTR);
803 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
804 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
805 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
806 struct lu_fid op_fid4; /* to the operation locks. */
807 u32 op_mds; /* what mds server open will go to */
809 struct lustre_handle op_handle;
813 struct lmv_stripe_md *op_mea1;
814 struct lmv_stripe_md *op_mea2;
815 __u32 op_suppgids[2];
822 /* iattr fields and blocks. */
823 struct iattr op_attr;
824 loff_t op_attr_blocks;
825 __u64 op_valid; /* OBD_MD_* */
826 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
828 enum md_op_flags op_flags;
830 /* Various operation flags. */
831 enum mds_op_bias op_bias;
833 /* used to transfer info between the stacks of MD client
834 * see enum op_cli_flags */
835 enum md_cli_flags op_cli_flags;
837 /* File object data version for HSM release, on client */
838 __u64 op_data_version;
839 struct lustre_handle op_lease_handle;
841 /* File security context, for creates. */
842 const char *op_file_secctx_name;
843 void *op_file_secctx;
844 __u32 op_file_secctx_size;
846 /* default stripe offset */
847 __u32 op_default_stripe_offset;
851 /* Used by readdir */
852 unsigned int op_max_pages;
857 int (*md_blocking_ast)(struct ldlm_lock *lock,
858 struct ldlm_lock_desc *desc,
859 void *data, int flag);
862 struct md_enqueue_info;
863 /* metadata stat-ahead */
864 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
865 struct md_enqueue_info *minfo,
868 struct md_enqueue_info {
869 struct md_op_data mi_data;
870 struct lookup_intent mi_it;
871 struct lustre_handle mi_lockh;
872 struct inode *mi_dir;
873 struct ldlm_enqueue_info mi_einfo;
874 md_enqueue_cb_t mi_cb;
879 struct module *o_owner;
880 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
881 void *karg, void __user *uarg);
882 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
883 __u32 keylen, void *key, __u32 *vallen, void *val);
884 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
885 __u32 keylen, void *key,
886 __u32 vallen, void *val,
887 struct ptlrpc_request_set *set);
888 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
889 int (*o_precleanup)(struct obd_device *dev);
890 int (*o_cleanup)(struct obd_device *dev);
891 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
892 int (*o_postrecov)(struct obd_device *dev);
893 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
895 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
896 /* connect to the target device with given connection
897 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
898 * granted by the target, which are guaranteed to be a subset of flags
899 * asked for. If @ocd == NULL, use default parameters. */
900 int (*o_connect)(const struct lu_env *env,
901 struct obd_export **exp, struct obd_device *src,
902 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
904 int (*o_reconnect)(const struct lu_env *env,
905 struct obd_export *exp, struct obd_device *src,
906 struct obd_uuid *cluuid,
907 struct obd_connect_data *ocd,
909 int (*o_disconnect)(struct obd_export *exp);
911 /* Initialize/finalize fids infrastructure. */
912 int (*o_fid_init)(struct obd_device *obd,
913 struct obd_export *exp, enum lu_cli_type type);
914 int (*o_fid_fini)(struct obd_device *obd);
916 /* Allocate new fid according to passed @hint. */
917 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
918 struct lu_fid *fid, struct md_op_data *op_data);
921 * Object with @fid is getting deleted, we may want to do something
924 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
925 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
926 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
927 __u64 max_age, struct ptlrpc_request_set *set);
928 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
930 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
932 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
934 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
936 int (*o_preprw)(const struct lu_env *env, int cmd,
937 struct obd_export *exp, struct obdo *oa, int objcount,
938 struct obd_ioobj *obj, struct niobuf_remote *remote,
939 int *nr_pages, struct niobuf_local *local);
940 int (*o_commitrw)(const struct lu_env *env, int cmd,
941 struct obd_export *exp, struct obdo *oa,
942 int objcount, struct obd_ioobj *obj,
943 struct niobuf_remote *remote, int pages,
944 struct niobuf_local *local, int rc);
945 int (*o_init_export)(struct obd_export *exp);
946 int (*o_destroy_export)(struct obd_export *exp);
948 int (*o_import_event)(struct obd_device *, struct obd_import *,
949 enum obd_import_event);
951 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
952 enum obd_notify_event ev);
954 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
955 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
958 int (*o_quotactl)(struct obd_device *, struct obd_export *,
959 struct obd_quotactl *);
961 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
964 int (*o_pool_new)(struct obd_device *obd, char *poolname);
965 int (*o_pool_del)(struct obd_device *obd, char *poolname);
966 int (*o_pool_add)(struct obd_device *obd, char *poolname,
968 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
970 void (*o_getref)(struct obd_device *obd);
971 void (*o_putref)(struct obd_device *obd);
973 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
974 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
975 * Also, add a wrapper function in include/linux/obd_class.h. */
980 struct mdt_body *body;
981 struct lu_buf layout;
982 struct lmv_stripe_md *lmv;
983 #ifdef CONFIG_FS_POSIX_ACL
984 struct posix_acl *posix_acl;
988 struct md_open_data {
989 struct obd_client_handle *mod_och;
990 struct ptlrpc_request *mod_open_req;
991 struct ptlrpc_request *mod_close_req;
992 atomic_t mod_refcount;
996 struct obd_client_handle {
997 struct lustre_handle och_fh;
998 struct lu_fid och_fid;
999 struct md_open_data *och_mod;
1000 struct lustre_handle och_lease_handle; /* open lock for lease */
1005 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1007 struct lookup_intent;
1011 /* Every operation from MD_STATS_FIRST_OP up to and including
1012 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1013 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1014 * Operations after MD_STATS_LAST_OP are excluded from stats.
1015 * There are a few reasons for doing this: we prune the 17
1016 * counters which will be of minimal use in understanding
1017 * metadata utilization, we save memory by allocating 15
1018 * instead of 32 counters, we save cycles by not counting.
1020 * MD_STATS_FIRST_OP must be the first member of md_ops.
1022 #define MD_STATS_FIRST_OP m_close
1023 int (*m_close)(struct obd_export *, struct md_op_data *,
1024 struct md_open_data *, struct ptlrpc_request **);
1026 int (*m_create)(struct obd_export *, struct md_op_data *,
1027 const void *, size_t, umode_t, uid_t, gid_t,
1028 cfs_cap_t, __u64, struct ptlrpc_request **);
1030 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1031 const union ldlm_policy_data *, struct md_op_data *,
1032 struct lustre_handle *, __u64);
1034 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1035 struct ptlrpc_request **);
1037 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1038 struct lookup_intent *,
1039 struct ptlrpc_request **,
1040 ldlm_blocking_callback, __u64);
1042 int (*m_link)(struct obd_export *, struct md_op_data *,
1043 struct ptlrpc_request **);
1045 int (*m_rename)(struct obd_export *, struct md_op_data *,
1046 const char *, size_t, const char *, size_t,
1047 struct ptlrpc_request **);
1049 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1050 size_t , struct ptlrpc_request **);
1052 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1053 struct ptlrpc_request **);
1055 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1056 struct md_callback *cb_op, __u64 hash_offset,
1057 struct page **ppage);
1059 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1060 struct ptlrpc_request **);
1062 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1063 u64, const char *, const char *, int, int, int, u32,
1064 struct ptlrpc_request **);
1066 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1067 u64, const char *, const char *, int, int, int,
1068 struct ptlrpc_request **);
1070 int (*m_intent_getattr_async)(struct obd_export *,
1071 struct md_enqueue_info *);
1073 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1074 struct lu_fid *, __u64 *bits);
1076 #define MD_STATS_LAST_OP m_revalidate_lock
1078 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1079 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1081 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1082 struct ptlrpc_request **);
1084 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1086 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1087 struct obd_export *, struct obd_export *,
1088 struct lustre_md *);
1090 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1092 int (*m_merge_attr)(struct obd_export *,
1093 const struct lmv_stripe_md *lsm,
1094 struct cl_attr *attr, ldlm_blocking_callback);
1096 int (*m_set_open_replay_data)(struct obd_export *,
1097 struct obd_client_handle *,
1098 struct lookup_intent *);
1100 int (*m_clear_open_replay_data)(struct obd_export *,
1101 struct obd_client_handle *);
1103 int (*m_set_lock_data)(struct obd_export *,
1104 const struct lustre_handle *, void *, __u64 *);
1106 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1107 const struct lu_fid *, enum ldlm_type,
1108 union ldlm_policy_data *, enum ldlm_mode,
1109 struct lustre_handle *);
1111 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1112 union ldlm_policy_data *, enum ldlm_mode,
1113 enum ldlm_cancel_flags flags, void *opaque);
1115 int (*m_get_fid_from_lsm)(struct obd_export *,
1116 const struct lmv_stripe_md *,
1117 const char *name, int namelen,
1118 struct lu_fid *fid);
1119 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1120 const union lmv_mds_md *lmv, size_t lmv_size);
1123 static inline struct md_open_data *obd_mod_alloc(void)
1125 struct md_open_data *mod;
1129 atomic_set(&mod->mod_refcount, 1);
1133 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1134 #define obd_mod_put(mod) \
1136 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1137 if ((mod)->mod_open_req) \
1138 ptlrpc_req_finished((mod)->mod_open_req); \
1139 OBD_FREE_PTR(mod); \
1143 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1144 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1145 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1147 /* return 1 if client should be resend request */
1148 static inline int client_should_resend(int resend, struct client_obd *cli)
1150 return atomic_read(&cli->cl_resends) ?
1151 atomic_read(&cli->cl_resends) > resend : 1;
1155 * Return device name for this device
1157 * XXX: lu_device is declared before obd_device, while a pointer pointing
1158 * back to obd_device in lu_device, so this helper function defines here
1159 * instead of in lu_object.h
1161 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1163 return lu_dev->ld_obd->obd_name;
1166 static inline bool filename_is_volatile(const char *name, size_t namelen,
1172 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1175 /* caller does not care of idx */
1179 /* volatile file, the MDT can be set from name */
1180 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1181 /* if no MDT is specified, use std way */
1182 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1184 /* test for no MDT idx case */
1185 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1186 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1190 /* we have an idx, read it */
1191 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1192 *idx = simple_strtoul(start, &end, 16);
1194 * no digit, no trailing :, negative value
1196 if (((*idx == 0) && (end == start)) ||
1197 (*end != ':') || (*idx < 0))
1202 /* bad format of mdt idx, we cannot return an error
1203 * to caller so we use hash algo */
1204 CERROR("Bad volatile file name format: %s\n",
1205 name + LUSTRE_VOLATILE_HDR_LEN);
1209 static inline int cli_brw_size(struct obd_device *obd)
1211 LASSERT(obd != NULL);
1212 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1215 /* when RPC size or the max RPCs in flight is increased, the max dirty pages
1216 * of the client should be increased accordingly to avoid sending fragmented
1217 * RPCs over the network when the client runs out of the maximum dirty space
1218 * when so many RPCs are being generated.
1220 static inline void client_adjust_max_dirty(struct client_obd *cli)
1223 if (cli->cl_dirty_max_pages <= 0)
1224 cli->cl_dirty_max_pages = (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024)
1227 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1228 cli->cl_max_pages_per_rpc;
1230 if (dirty_max > cli->cl_dirty_max_pages)
1231 cli->cl_dirty_max_pages = dirty_max;
1234 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1235 cli->cl_dirty_max_pages = totalram_pages / 8;
1238 #endif /* __OBD_H */